A non-exponential transmittance model for volumetric scene representations

We introduce a novel transmittance model to improve the volumetric representation of 3D scenes. The can represent opaque surfaces in light transport framework. Volumetric representations are useful for complex scenes, and become increasingly popular level detail scene reconstruction. traditional exponential found cannot capture correlations visibility across volume elements. When representing as density, this leads both bloating silhouettes leaking artifacts. By introducing parametric non-exponential model, we able approximate these correlation effects significantly accuracy appearance Our continuum between linear that exhibit encountered participating media unstructured geometries. This covers large part spectrum geometric structures In order handle spatially varying effects, further extend theory heterogeneous model. is compact storage computationally efficient evaluation reverse-mode gradient computation. Applying our optimization algorithms yields significant improvements quality. show relevant applications, such prefiltering, image-based reconstruction using differentiable rendering, neural representations, compare it conventional